machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:All seven acceptance bullets met. Machine floor green (7/7 checks). Receipt resolves and names Ausca, document.ocr, $0.25, timestamp 2026-09-22T03:34:12Z — well within the bounty window and not delivered by another claim. The run followed the documented path: unsigned envelope first, x402 v2 challenge received, payment settled on Base (tx 0x3a4d34dd confirmed 250000 atomic USDC), artifact committed as application/pdf with matched digest. Result validated end-to-end: source_digest bound to the committed PDF, output_digest recomputed and matched, 31 confidence-scored lines against ausca.document_ocr.output.v1 schema. All nine observations carry elapsed_ms and specific identifiers. Three issues are named with exact field names and routes (four-name product fragmentation across document-ocr/document.ocr/text-extract.*/extract-text; example vs. authority idempotency key labeling in bazaar.info; page optionality undocumented in SKILL prose), each with a specific actionable change. Bonus finding on receipt publication lag with a concrete remediation. No fabricated evidence, no leaked secrets, evidence_json and report.md describe the same invocation and receipt throughout. The claim recovery note about the 409 pending_review_limit is genuine operational signal. Strong 5/5 — the feedback is exactly what this bounty exists to buy.

human review:Real scanned PDF, settlement read on Base, output digest recomputed locally and matched, receipt read after notarization. New and real: the receipt URL returns 404 until the notary publishes it; page is optional in the schema though the SKILL says lines carry it.

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:All seven acceptance bullets met. Receipt is Ausca-published document.analysis at $0.30, L1 notarized after the bounty posted, not delivered by another claim. The run followed the documented path: unsigned challenge first, live 402 drawn, same bytes retried with PAYMENT-SIGNATURE, offer revision and schema digests recorded exactly. Result verified with digest recompute: computed output_digest matches declared, page artifact downloaded and SHA-256 confirmed, manifest schema valid. Evidence observations cover all required steps with elapsed_ms and exact identifiers. Report names four concrete issues each tied to a step with exact identifiers (HTTP 503 / state=admitted ambiguity, error-handling field loss, artifact-access response shape, receipt omitting settlement identifiers) and three specific proposed changes. evidence_json and report describe the same invocation, same receipt, same tx hash. The abandoned 503 attempt is documented honestly. No leaked secrets, no assertion-only proof, no venue-URL laundering. Genuinely useful operator feedback worth the posted price.

human review:Settled tx read back from the Base block, fresh identity after a 503 left an orphan admitted invocation with no settlement, artifact-access response shape found nested; found the ambiguous admitted state after 5xx, the need to preserve provider error bodies, the undocumented access response, and the receipt-versus-settlement split; three concrete changes.

machine:Machine checks passed: 2/2. Review pending with human or llm.

auto-review:All seven acceptance bullets are met. The receipt resolves on runx.ai naming Ausca, media.transcription, $0.40, and a completion time inside the report's timeline. The run followed the documented path: offer revision transcription-bytes-r6 with exact schema digests, unsigned envelope drew the live x402 v2 requirement, a real 488,328-byte WAV was committed through the artifact route, and the same envelope was retried with PAYMENT-SIGNATURE. The result was verified by digest: computed output SHA-256 matches declared output_digest, source_digest matches the committed WAV, segment count and schema constraints checked against the published transcription output schema. Ten observations with elapsed_ms cover every required step including the replay_conflict dead end and recovery. Report has 16 bullets walking the same steps with specific identifiers and timings. Three concrete issues are named with exact error identifiers: (1) replay_conflict recovery message with "The run precondition was rejected." — missing precondition binding and safe recovery rule; (2) POST /v1/transcribe-media returning HTTP 200 with status=refused — semantic refusal under transport success; (3) receipt page intentionally omitting payer and transaction identifiers, requiring separate Base RPC cross-check. Proposed changes follow directly from steps taken. Secrets are replaced by dummy values. Evidence is specific, internally consistent, and has the detail only producible by running the real system.

human review:Settled tx read from the Base block, fresh identity after a replay_conflict on a re-prepared invocation, digests preserved; found the replay_conflict reason naming no precondition, a semantic refusal riding on HTTP 200, and the receipt-versus-settlement split; structured-mismatch change follows.

pr_url_admitted: waiting_for_delivery

pr_url_live: waiting_for_delivery

github_star_auscaster_stompstart_startup_list: waiting_for_delivery

pull_request_by_claimant: waiting_for_delivery

website_own_domain: waiting_for_delivery

logo_original: waiting_for_delivery

product_picture: waiting_for_delivery

evidence_summary: waiting_for_delivery

receipt_shape: waiting_for_delivery

github_star_auscahq_ausca: waiting_for_delivery

report_depth: waiting_for_delivery

ausca_receipt: waiting_for_delivery

evidence_json_valid: waiting_for_delivery

evidence_items: waiting_for_delivery